Four years ago, my uncle sent a mass e-mail to me and a bunch of other relatives asking us to make election prediction maps and explain our reasoning, and that he would send a token prize of $1 to whoever was closest. I figured, if you want to make a specific prediction of outcomes rather than just repeat the probabilities that a model spits out, you should start with the probabilistic map and then consider a polling error one way or the other. I produced two maps: one representing Trump outperforming his polls by a few points and one representing Clinton outperforming hers. My “Trump outperforms” map was indeed the closest to the actual result, but my uncle decided we all still lost because none of us predicted Trump would win — a pretty normal response to pretty normal polling errors, even after years to reflect.
The takeaway lessons should, I think, be that even small differences in the polls can have large effects on our bizarre electoral college system and that we should be prepared for statistically-noisy errors or late shifts rather than uniform ones, things unique to a place or a demographic that cause unexpected but understandable results. Or possibly it’s that I shouldn’t enter contests that have unclear rules.
But I’d like to take another spin at a prediction with a very different approach: assuming the polls are basically right for likely voters, and that a potential Democratic benefit from unlikely voters will be competing with a Republican benefit from Democrats voting by mail more and therefore getting their votes disqualified more.
States of a wide variety of regions and political leanings have in some way changed their voting laws to support this surge in early and mail voting due to the pandemic. The high early turnout, both in person and by mail, skews even more Democratic than usual. This is certainly largely because they’re more likely to expect the pandemic and voter suppression to interfere with the voting process and are trying to avoid those obstacles. But turnout in general is expected to be exceptionally high in this election, and the expansion of early and absentee voting is encouraging people who have never voted before to do so. You don’t get the highest turnout in over a century with only “likely voters”, and unlikely voters don’t often show up in polls. Perhaps I’m being too confident about high turnout favoring Democrats, but there might even be some polling evidence of newly-registered and registered-but-infrequent voters going for Biden in the swing states.
But while voters take their states up on new vote by mail options, this shift is a double-edged sword: for millions of people, whether this is overall their first election or their fortieth, this is the first election where they’ve voted by mail, and that might mean a high rate of rejection due to honest confusion by voters. Election officials sounded the alarm back in July that changes to the voting laws might lead even experienced voters to cast invalid votes if they don’t know about, or just misunderstand, their states’ particular changes.
Both new and established policies have also been challenged in court — and with some policies getting alternately upheld or halted several times at various levels of the appeals process, I do not at all blame the average voter for not knowing whether a particular way to vote is legal for them. For a particularly dramatic example, look at Pennsylvania, where multiple challenges to the deadline for receiving ballots worked their way through the courts in October, so voters who voted (or wanted to vote) by mail close to the election date may well have not known whether their votes, which seemed legitimate at the time, would be counted. And still might not know until after Election Day.
The result of all of this? I strongly suspect that how states have adapted to early and mail voting will play an outsized influence, and that fear of voter suppression might actually be driving a Democratic-leaning increase in turnout among people determined to fight the barriers to voting. So, like in our family contest four years ago, I’m going to start with FiveThirtyEight’s projected vote share for each state (as of 6PM Eastern time on November 2nd), but instead of adjusting it based on possible error, I will make some
arbitrary and not well-researched small and plausible-sounding adjustments, based on the following heuristics:
- Low experience with voting by mail means a likely botched implementation of vote by mail this time around, and since Biden voters are disproportionately voting by mail, that means failures in distributing or counting ballots will hurt him more. States with < 10% of 2016 votes cast by mail will be adjusted toward Trump.
- Exceptionally high early voting turnout probably indicates that voters are actively working around barriers to voting. This will tend to overwhelm factors that reduce turnout, whether those factors are intentional or not, and this may represent new or inconsistent voters who may not show up in polls. States not given the penalty above will be adjusted toward Biden, as will those that weren’t already doing universal-vote-by-mail before this election whose 2020 early turnout is at least ⅔ of their 2016 total turnout. States whose early turnout already exceeds 2016’s total will be adjusted more. (I set this rule before it became clear that only Texas and Hawaii would achieve this benchmark; I swear this isn’t special pleading.)
- Last-minute changes will tend to confuse people, reducing turnout and increasing invalid votes. I included most, but not all, of the changes and lawsuits that showed up in FiveThirtyEight’s coverage of how to vote and the live and recent issues about it (again, to be clear, I’m not pretending this is a scientific effort here — it’s more a way to force my guesses to be consistent with each other). States which recently made changes that seem to me like they would make people believe conflicting things about how to vote will generally be adjusted toward Trump.
- States will tend to reject large numbers of good-faith votes if they use Election Day as the deadline by which ballots must be received rather than as the deadline by which they must be mailed. States which have such requirements will be adjusted toward Trump.
“Toss-up” tan: margin of victory < 1 point (I doubt we’ll have complete results by the end of the week, except possibly Georgia, which just barely tilts Biden here)
“Lean” light shades: margin 1-4.8 points (would flip with an average-sized state-level polling error)
“Likely” medium shades: margin 4.8-9 points (would flip with a large but not unprecedented error)
“Safe” dark shades: margin > 9 points (something very weird would have to happen for this to be wrong)
Most states are barely affected at all under my scenario, at least in terms of which candidate they’re likely to go to, as conflicting effects cancel each other out. Alaska and Minnesota, for example, have exactly the same adjustments in each direction. But you can see a few notable differences between this projection and just the polls. Arizona, for example, with very high early turnout and familiarity with voting by mail, shifts a few points bluer, from “lean Biden” to “likely”. Pennsylvania, meanwhile, has very little past vote-by-mail experience and has been targeted by several attempts to change the deadline for receiving mail ballots, and shifts redder by the same amount. Texas outweighs the confusion of its policy change shenanigans with a truly astonishing early turnout and a court rejecting an attempt to throw out over a hundred thousand early votes to become one of three states with an estimated margin of less than 1 percentage point between the candidates.
The biggest question here for me is Pennsylvania. If it is as close as I suspect here, that may also not get us complete results by the end of the week. And although it still leans Biden, as I alluded to earlier, there is still an effort underway to blatantly disregard the election results by throwing out ballots that seemed to have been legally cast at the time. That, or even a small polling error, or both, would push it into Trump’s column — and as FiveThirtyEight likes to point out, polling error in one state predicts polling error in others with similar demographics and in the same region, which means Pennsylvania is an outsized indicator of what might happen for the election as a whole.
Of course, this is just one scenario. Biden could systematically underperform precisely because of widespread voting by mail. Or pollsters could be overcorrecting for demographic groups who are hard to poll, which may look similar to that. Or paramilitary intimidation of voters, or the fear that it might happen, could drive down turnout in certain states. Or pollsters could have gone too far in their education weighting adjustments from last election and Trump could underperform among voters without college degrees. Or nothing particularly dramatic could happen but I could just be completely wrong about gains. Or polling errors might go in different directions in different states, or voters who wouldn’t normally turn out might favor different candidates in different states, neither of which I can even guess at how to map. Overall, although my scenario seems plausible to me, I acknowledge that it relies on speculation, and I caution you that there is still a world of other possible election outcomes. Including the polls just… being very close to correct. Again.
In the Our Long National Nightmare Facebook group, we’re challenging our readers to try their hand at predicting the outcome themselves, with the prize being an as-of-yet unreleased t-shirt (you can see our released goods in the Pyramid Consumer Experience).